De-Noising and Analysis of Simulated Intensity Distributions from Non-Imaging Systems

Ola Wassvik

Handledare: Ulla Holst

Centre for Mathematical Sciences
Mathematical Statistics
Lund Institute of Technology,
Lund University,
2002:E12


Abstract
Non-parametric regression methods are often used in statistics. The bandwidth selection methods are often crucial for the outcome. In this thesis we are suggesting a method for choosing the bandwidths adaptively from the data. Thus we get a method that requires minimal input from the operator.
This regression method is then used for estimating light distributions in the design process of a machine vision illumination system.
Simulated and measured datasets are used to study the estimation procedure. We study the restrictions and future uses for the procedure. Parameter restrictions are found and we suggest continued improvements.